Game Testing Methodology
We used Fraps to measure the frame rates for each of today’s games, and will focus purely on 1920x1080, a popular resolution well within the Radeon HD 7970’s capabilities. We’re most concerned about the highest playable settings, but also ran less demanding graphics details to better flesh out CPU limitations and scaling.
We're plotting Average and Minimum Frames Per Second (FPS) in bar charts, and Frame Rate Over Time in line graphs. But for this CPU-related story, we are not logging Frame Times. Nvidia’s Frame Capture Analysis Tools (FCAT) are in heavy rotation across two Tom's Hardware labs, and are an integral part of multi-GPU stories like AMD Radeon HD 7990: Eight Games And A Beastly Card For $1,000. It's not as imperative for single-GPU coverage though, since the data recorded by Fraps and FCAT come much closer to being the same. Nevertheless, I chose not to log frame times with Fraps until a few other issues can be investigated. More than anything, I wanted to be sure that any frame time variance recorded in Fraps was a direct a result of the processor, and would occur during normal gaming, rather than just overhead from the software collecting data.
Subjectively, I’ll come out and say that there wasn't a single case where acceptable performance was hurt by noticeably inconsistent frame times simply by swapping CPUs. I can’t rule out the likelihood that frame latency might have impacted the fluidity of already sub-par frame rates, though.
This phenomenon did occur in Battlefield 3, but since it was equally evident with every processor, and carried over to both platforms, it was most likely attributable to the graphics card and drivers. At the Ultra detail preset, frame rates never dropped below 60 FPS, and both averages and minimums were within 3 FPS of each other for all tested processors. Even so, the game still didn’t appear fully smooth. I chose not to use Battlefield 3 data in today's story simply because there was almost no CPU scaling in our normal 90-second single-player benchmark routine. And for a story like this, you're going to worry more about the multiplayer experience anyway.
But don’t worry. On the following pages, we’ll explore data for eight other games, including many of the latest heavy-hitters. If anything, we could argue this test suite is overly brutal, but the idea is to push each processor to its limit, gauging current, and, if possible, future demands.